I know there’s a large literature on this, but I am no expert. One way to check for consistency is to ask whether there is an ordering of the justices on the real line such that for almost all decisions, there is an x such that justices to the left of x vote “yes” and those to the right of x vote “no”, or vice versa. I believe this is the case, but I’d also like to see some numbers. - Cris

On Jul 3, 2018, at 10:12 PM, Keith F. Lynch <kfl@KeithLynch.net> wrote:

I hope I can be forgiven a brief digression into politics on this, the US's 242nd Independence Day.

In response to the pending resignation of one of the nine current Supreme Court justices, charts showing all nine, in order of their alleged position on the supposed political spectrum, have been in the news lately.

I've long thought that the "left" and "right" are more like random and shifting assemblages of ideas than any sort of coherent, persistent, or self-consistent political, legislative, legal, economic, or judicial philosophies. I usually can't tell who is what. And people who know me strongly disagree with each other on where on the political spectrum I am (except that nobody thinks I'm moderate).

Can we use math to determine whether the idea of a single one-dimensional political spectrum has any utility in explaining the behavior of Supreme Court justices, senators, U.S. states, etc.? Can we arrange the justices (or senators, etc.) in a line such that for every yes-no decision they made, there's a point which everyone who voted yes is on one side of and everyone who voted no is on the other side of? The point can be in a different place for each such decision, and yes and no can swap sides. But, for instance, if justices A, B, and C are in that order, A and C should never agree with each other except when B agrees with both of them.

The answer is almost certainly no. But how close can we get? Can we get it to work for 99% of their decisions? For 90%?

What's an efficient algorithm for finding out? With just nine, it's practical, with a modern computer, to try all 181440 (9!/2) possible orders. But with 50 states, 100 senators, or 535 members of congress, that would obviously be intractable.

Some claim that the two extremes of the spectrum are identical, or close to it. We could test that by making the spectrum a circle, and seeing if there's always a line that will divide everyone who voted one way from everyone who voted the other. (A different line each time, but with the justices (or senators, etc.) always in the same order around the circle.)

Or we could try a two-dimensional spectrum. Place the justices on a plane such that for every decision a line exists that divides the yes votes from the no votes. Again, the spectrum may wrap around in one (a cylinder) or both (a torus) dimensions.

Similarly with higher dimensions. Of course with N voters, if you choose N-1 dimensions you can always find an N-2-plane that divides the yes votes from the no votes, even if they're completely random. Just as any N points can be perfectly, but uselessly, fit by some order N-1 polynomial.

But I'm way over my pay grade, even for two dimensions. Just how many "orderings" of N voters in a plane are there, for the purposes of finding whether such a line exists?

Has anyone already done such work, especially with real-world data? Also, where can I find complete machine-readable real-world voting records for justices, senators, states (in presidential elections), etc.? Thanks.

_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun

Check out Minsky's (1950's?) work on "Perceptrons". He was trying to show how weak they were at certain decision tasks. At 09:12 PM 7/3/2018, Keith F. Lynch wrote:

I hope I can be forgiven a brief digression into politics on this, the US's 242nd Independence Day.

In response to the pending resignation of one of the nine current Supreme Court justices, charts showing all nine, in order of their alleged position on the supposed political spectrum, have been in the news lately.

I've long thought that the "left" and "right" are more like random and shifting assemblages of ideas than any sort of coherent, persistent, or self-consistent political, legislative, legal, economic, or judicial philosophies. I usually can't tell who is what. And people who know me strongly disagree with each other on where on the political spectrum I am (except that nobody thinks I'm moderate).

Can we use math to determine whether the idea of a single one-dimensional political spectrum has any utility in explaining the behavior of Supreme Court justices, senators, U.S. states, etc.? Can we arrange the justices (or senators, etc.) in a line such that for every yes-no decision they made, there's a point which everyone who voted yes is on one side of and everyone who voted no is on the other side of? The point can be in a different place for each such decision, and yes and no can swap sides. But, for instance, if justices A, B, and C are in that order, A and C should never agree with each other except when B agrees with both of them.

The answer is almost certainly no. But how close can we get? Can we get it to work for 99% of their decisions? For 90%?

What's an efficient algorithm for finding out? With just nine, it's practical, with a modern computer, to try all 181440 (9!/2) possible orders. But with 50 states, 100 senators, or 535 members of congress, that would obviously be intractable.

Some claim that the two extremes of the spectrum are identical, or close to it. We could test that by making the spectrum a circle, and seeing if there's always a line that will divide everyone who voted one way from everyone who voted the other. (A different line each time, but with the justices (or senators, etc.) always in the same order around the circle.)

Or we could try a two-dimensional spectrum. Place the justices on a plane such that for every decision a line exists that divides the yes votes from the no votes. Again, the spectrum may wrap around in one (a cylinder) or both (a torus) dimensions.

Similarly with higher dimensions. Of course with N voters, if you choose N-1 dimensions you can always find an N-2-plane that divides the yes votes from the no votes, even if they're completely random. Just as any N points can be perfectly, but uselessly, fit by some order N-1 polynomial.

But I'm way over my pay grade, even for two dimensions. Just how many "orderings" of N voters in a plane are there, for the purposes of finding whether such a line exists?

Has anyone already done such work, especially with real-world data? Also, where can I find complete machine-readable real-world voting records for justices, senators, states (in presidential elections), etc.? Thanks.